The forecasting potential of complex models

The nature and use of “complex” models for forecasting and policy simulation are analyzed on theoretical and empirical-performance grounds. The analysis suggests that while the accuracy of complex models in forecasting trends in such fields as economic and energy is, and will remain, undistinguished, complex models' special virtues of preserving counter-intuitive results and representing subsystem interdepedence could be used to better advantage than current practice permits. Suggestions for such improvements, through more diversified model structures, micro-process models in addition to the typical macro models, a mix of mechanically- and judgmentally-operated models, and the modeling of policy response, are reviewed.